Signaling asset price bubbles with time-series methods

This paper provides an early warning indicator for bubbles in financial markets. The indicator is based on traditional unit root tests, more precisely on the augmented Dickey-Fuller test and may be used in a repeated manner with rolling samples. The performance of the indicator is tested extensively via Monte Carlo simulations and comparisons of the results with the most powerful standard (stability) tests. The new indicator seems to be more robust and to have more power than the standard tests. In empirical application to US stock market data for 1871-2010, the new indicator signals most of the consensus bubbles and gives warning signals well ahead of the crash, in most cases as early as 12 months ahead. The indicator also signals most of the 'negative bubbles' before their turning points. The author would like to thank Matti Viren, Esa Jokivuolle, Jouko Vilmunen, Pentti Saikkonen, Heikki Kauppi and Ari Hyytinen for their comments at various stages of this work. I would also like to thank Nina Björklund and Tarja Yrjölä for research assistance.